Our project's code is based on ClementPinard's SfmLearner-Pytorch https://github.com/ClementPinard/SfmLearner-Pytorch which implements the system described in the paper:
Unsupervised Learning of Depth and Ego-Motion from Video
Tinghui Zhou, Matthew Brown, Noah Snavely, David G. Lowe
In CVPR 2017 (Oral).
See the project webpage for more details.
To accomplish the goal of our project, we add newly-designed architectures (in "models" and "mono2net" folders), new loss functions (in "loss_functions.py") and new training schemes, etc, directly into the original scripts. We also modified required functional scripts for our use.